error installing libmesh on a HPC cluster (petsc-3.14.2) #17040
-
|
Dear all, |
Beta Was this translation helpful? Give feedback.
Replies: 8 comments 6 replies
-
|
Dear all, |
Beta Was this translation helpful? Give feedback.
-
|
Attached the whole build.log (>4 MB) file for libmesh in case it could be of any help (the config.log is as for the one above and it looks OK to me at least) |
Beta Was this translation helpful? Give feedback.
-
|
Hi Mauro Is this fixed? I ll tag people who may be able to help if not |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for your reply @GiudGiud. |
Beta Was this translation helpful? Give feedback.
-
|
I saw this 28 days ago, and I still don't have an idea... @roystgnr or @lindsayad ? |
Beta Was this translation helpful? Give feedback.
-
|
@fdkong was working on push_parallel_vectors lately he might have an idea too |
Beta Was this translation helpful? Give feedback.
-
|
It could be an "old" compiler. Could you should the result of "mpicxx -show"? |
Beta Was this translation helpful? Give feedback.
-
|
@fdkong I am loading gcc 7.2 on openmpi 4. Here is the show on the mpicxx: Should I try to see whether I can upgrade? |
Beta Was this translation helpful? Give feedback.
-
|
We should definitely still be supporting 7.2; if not then that's something that we should fix or workaround on our part. But it couldn't hurt to try a newer compiler to confirm or rule that out. |
Beta Was this translation helpful? Give feedback.
-
|
Oh, wait! Just looked at your build.log. We've seen this once before - the trouble is likely that we've started requesting more recent C++ standards support by default, but some older compilers accept the flags to support newer standards, pass our configure-time tests with those flags on, but then fail to actually compile our code. Try forcing an older standard by configuring libMesh with |
Beta Was this translation helpful? Give feedback.
-
|
We really need @jwpeterson to work his m4 C++ standard testing magic |
Beta Was this translation helpful? Give feedback.
-
|
The trouble here is that I'm not even sure how we'd write a test for this problem. It's not a missing feature that we're running into; it'seems to be a compiler bug that gets triggered by a very complex piece of code. It'd be a ton of work to boil it down into a test small enough for an autoconf macro that still triggers the same problem. I'd say we should just downgrade the level of C++ we ask for, since we don't need C++17 yet, but I've been hoping we could actually upgrade to need (and thus to be able to internally use) C++17 later this year, now that even the conservative RHEL distributions default to gcc 8. But if people are still using gcc 7.2 and if it's responsible for this build failure then perhaps I'm being too optimistic? gcc 7 is supposed to have full C++17 support too, in theory, but in practice here we are. |
Beta Was this translation helpful? Give feedback.
-
|
@roystgnr and all other: Thanks! I really had some terrible time trying to figure out what I was doing wrong ... |
Beta Was this translation helpful? Give feedback.
-
|
After upgrading to gcc 9.2 libmesh install without any problem. @roystgnr: your guess was right. Though I thought I did try to force the 11 standards before, but surely I did mix things up. Thanks for the support. |
Beta Was this translation helpful? Give feedback.
@fdkong I am loading gcc 7.2 on openmpi 4. Here is the show on the mpicxx:
g++ -I/cluster/openmpi/gcc-7.2.0/openmpi-4.0.1/include -pthread -Wl,-rpath -Wl,/cluster/openmpi/gcc-7.2.0/openmpi-4.0.1/lib -Wl,--enable-new-dtags -L/cluster/openmpi/gcc-7.2.0/openmpi-4.0.1/lib -lmpi
Should I try to see whether I can upgrade?